Google DeepMind and Singapore’s Government Technology Agency (GovTech) have built an AI model that spots scam conversations directly on a smartphone, DeepMind said on 14 October 2026 [1]. The tool has not been rolled out to the public yet: DeepMind did not give a launch date or say which apps will carry it.
The problem
According to DeepMind, scams “make up over half of Singapore’s reported crimes” and have caused “more than S$4 billion in losses since 2019” [1]. DeepMind says scams have shifted from simple phishing to social engineering, where criminals build trust through convincing conversations before asking for money, and that such scams “now account for up to 80% of all losses” [1].
How it works
The model is based on Gemma, Google’s family of small open AI models, fine-tuned on local scam conversations, from people posing as public bodies, banks and local figures to scams trending in the region [1].
It runs on the phone itself rather than in the cloud. DeepMind explains why: many messaging apps are end-to-end encrypted, so a company’s servers cannot read the messages to check them [1]. Doing the check on the device keeps conversations private.
To save battery, it works in two stages [1]:
- A small, fast classifier screens incoming messages for anything suspicious.
- Only if something looks off does the fine-tuned Gemma model read a larger part of the conversation to confirm a scam or dismiss a false alarm.
If it spots manipulation, it shows a warning on screen telling the user to pause before sharing money or sensitive details [1].
DeepMind says the model was trained in GovTech’s own cloud environment using anonymised data from scam reports and intelligence feeds, with no private or personally identifiable information [1].
“Fighting scams shouldn’t mean sacrificing personal privacy,” said Mark Chen, Director of Trust and Safety at GovTech Singapore. “This allows us to catch highly targeted, local scam attempts in real time while keeping personal communications completely private.” [1]
What this means
- It is not on phones yet. The work was presented at Singapore International Cyber Week 2026. DeepMind says messaging platforms and phone software makers could build these models into their products, but none have been named [1].
- Privacy and protection together. Checking messages on the device avoids sending private chats to a server.
- No accuracy figures yet. DeepMind has not published how often the model catches scams or raises false alarms.
Sources
- Google DeepMind (JJ Geewax and Miguel de Andrés), “Google DeepMind and Singapore Government announce new AI tool to protect citizens against scams”, 14 October 2026. https://deepmind.google/blog/on-device-ai-scam-detection-mddi/
